[core] Add physical search and full-text archive primitives

[JingsongLi]

Summary

Introduce the reusable physical-position search model and immutable full-text archive primitives needed by primary-key table indexes. This isolates the storage and ranking foundation before exposing primary-key full-text configuration or query APIs.

Changes

• Add shared physical search positions, scored results, deterministic ranking, and position-reader abstractions, then adapt the existing primary-key vector path to use them.
• Extend source-backed primary-key index metadata to retain ordered source files and an optional definition fingerprint.
• Add full-text archive state, projected data-file reading, archive building, and index-file handling with continuous row IDs across ordered sources.
• Cover metadata compatibility, null-row ordinals, multi-source archives, reader ownership, ranking, and vector compatibility with focused tests.

Testing

• JDK 8: compile paimon-core and its required modules
• Run 9 focused core test classes for source metadata, full-text archive primitives, indexed split reading, ranking, and vector compatibility
• Verify this is exactly one commit based on the target branch merge base

Notes

• This PR does not expose primary-key full-text table options, maintenance, or search APIs; those are layered in follow-up work.
• The complete follow-up implementation is currently tracked in [core] Support full-text search for primary-key tables #8648.
• No Python or index.type changes are included.

[leaves12138]

LGTM. Reviewed the physical search and full-text archive primitives, including the vector-index adaptations. The related paimon-core tests pass locally.